Up to date
LGAI-EXAONE exaone out 23 months ago
7.82B parameters, text in, text out, 32K tokens of context.
Verified 2026-10-25 Rank #471 in LLMs · score 77.3 method v1 Sources 3 Downloads 30d 247.5k · #235 of 619 tracked
Picked by rule from 1 tracked piece: the lab's own post, the most shared, a field report, the latest.
LG AI Research released three EXAONE 3.5 models including a 2.4B, 7.8B, and 32B variant.
Hugging Face does not say who downloads a model or with what. These are the proxies: API providers listing it, quantized builds, fine-tunes/adapters/merges, and Spaces that load it.
No API provider serves it, so its downloads are people running it themselves: 27 quantized builds and 10 Spaces.
Top Spaces running it: mayafree/Model-Genome-Korea ♥ 33
Weights only, an estimate: leave room for the KV cache. Method
About 4.7 GB at INT4, 9.4 GB at FP8, 19 GB at BF16: weights only, so leave room for the KV cache.
Estimate params × bytes per param × 1.2 · how we estimate
Weights, every dated article, and repo commits on one rail.
Teal is the weights (2024-12-01), amber triangles are 1 dated article (each links to it), grey ticks are repo commits.
319 derivatives on the Hub. Which build to run, by runtime: the most downloaded in each class.
From the Hugging Face model tree; quantized builds in teal, fine-tunes, merges and adapters in greys.
Everything else, folded: the full tables, every source.
| Benchmark | Value | Setting / notes | Reported by | Date |
|---|---|---|---|---|
| author-reported AlpacaEval · Chat and follow instructions | 54.2 | EXAONE 3.5 7.8B | model author | 2026-10-07 |
| author-reported Arena-Hard · Chat and follow instructions | 68.7 | EXAONE 3.5 7.8B | model author | 2026-10-07 |
| author-reported IFEval · Chat and follow instructions | 78.9 | EXAONE 3.5 7.8B | model author | 2026-10-07 |
| author-reported LiveBench · Reason through hard problems | 39.8 | EXAONE 3.5 7.8B | model author | 2026-10-07 |
| author-reported LogicKor | 9.08 | EXAONE 3.5 7.8B | model author | 2026-10-07 |
| author-reported MT-Bench · Chat and follow instructions | 8.29 | EXAONE 3.5 7.8B | model author | 2026-10-07 |
| author-reported MT-Bench · Chat and follow instructions | 7.96 | EXAONE 3.5 7.8B | model author | 2026-10-07 |
EXAONE-3.5-7.8B-Instruct is a 7.82B parameter bilingual instruction-tuned generative model developed by LG AI Research.[1][2] The model supports English and Korean languages and provides a context length of up to 32,768 tokens.[3][4]
Generated 2026-10-07 15:10 UTC by parasail-gemma-4-26b-a4b-it; each numeral points at the fact it rests on. Not a verdict.
Labels are evidence, not verdicts: measured a third-party leaderboard row for this model; author-reported a number the model author published in the card or eval files; structural a deterministic signal from the name, config or chat template, or a decision-model estimate. Each chip links to its source. All use cases.
Listed in LLMs
| Pipeline tag | text-generation |
|---|---|
| Library | transformers |
| Architectures | ExaoneForCausalLM |
| Model type | exaone |
| Dtype | float32 |
| Params by dtype | F32: 7,818,448,896 |
| Size band | M |
| MoE | no |
| Context length | 32,768 (config.json max_position_embeddings) |
| Languages | en, ko |
| Papers | arXiv:2412.04862 |
| License | exaone (other) |
| Hub snapshot | 2026-10-25 |
vLLM: supported · ExaoneForCausalLM · 7050ef881ff0835f6890a47db1859a24c65481d6SGLang: supported · ExaoneForCausalLM · 773cb938fe9484a8f7e7e46fb1b1fe96412c59ed
vLLM registry (7050ef881ff0835f6890a47db1859a24c65481d6) · SGLang registry (773cb938fe9484a8f7e7e46fb1b1fe96412c59ed) · Matched by architecture class name against each project's registry on main; 'supported' means the architecture is registered, not that this checkpoint was tested.
Weights-only memory for EXAONE-3.5-7.8B-Instruct: params × bytes/param × 1.2 overhead. KV cache and activations are extra. How this is computed.
| BF16 (2 bytes/param) | 19 GB · Single GPU (<24 GB) |
|---|---|
| FP8 (1 byte/param) | 9.4 GB · Single GPU (<24 GB) |
| INT4 (0.5 bytes/param) | 4.7 GB · Single GPU (<24 GB) |
Weights only (params × bytes/param × 1.2 overhead); KV cache and activations extra.
finetune: 136quantized: 27adapter: 156
| Variant | Relation | Quant | Params | Band | License | Author | Likes | Downloads |
|---|---|---|---|---|---|---|---|---|
| LGAI-EXAONE/EXAONE-Deep-7.8B | finetune | — | 7.82B | M | other | LGAI-EXAONE | 103 | 1.3M |
| lmstudio-community/EXAONE-3.5-7.8B-Instruct-GGUF | quantized | GGUF | — | — | other | lmstudio-community | 2 | 88.7k |
| Mungert/EXAONE-Deep-7.8B-GGUF | finetune | GGUF | — | — | other | Mungert | 5 | 29.4k |
| bartowski/EXAONE-3.5-7.8B-Instruct-GGUF | quantized | GGUF | — | — | other | bartowski | 16 | 27.3k |
| mradermacher/EXAONE-3.5-7.8B-Instruct-i1-GGUF | quantized | GGUF | — | — | other | mradermacher | 2 | 13.9k |
| mobilint/EXAONE-3.5-7.8B-Instruct | quantized | — | 419M | S | other | mobilint | 0 | 12.8k |
| OpenLLM-Korea/EXAONE-Deep-7.8B | finetune | — | 7.82B | M | other | OpenLLM-Korea | 0 | 4.9k |
| QuantFactory/EXAONE-Deep-7.8B-GGUF | finetune | GGUF | — | — | other | QuantFactory | 3 | 4.2k |
| mradermacher/EXAONE-3.5-7.8B-Instruct-GGUF | quantized | GGUF | — | — | other | mradermacher | 1 | 3.8k |
| second-state/EXAONE-3.5-7.8B-Instruct-GGUF | quantized | GGUF | — | — | other | second-state | 0 | 3.4k |
| MaziyarPanahi/EXAONE-3.5-7.8B-Instruct-GGUF | quantized | GGUF | — | — | — | MaziyarPanahi | 1 | 3k |
| hymmmm/exaone-3.5-7.8b-emt-awq | quantized | awq | 7.82B | M | other | hymmmm | 0 | 3k |
| tensorblock/EXAONE-3.5-7.8B-Instruct-GGUF | quantized | GGUF | — | — | other | tensorblock | 0 | 2.1k |
| gaianet/EXAONE-3.5-7.8B-Instruct-GGUF | quantized | GGUF | — | — | other | gaianet | 0 | 1.7k |
| ChanLumerico/EXAONE-3.5-7.8B-Instruct-Yaho | finetune | GGUF | 7.82B | M | other | ChanLumerico | 0 | 1.4k |
| Jongbin-kr/exaone_7b_lora_rtlcoder_ratio0.2 | adapter | — | — | — | — | Jongbin-kr | 0 | 1.2k |
| huggingface-KREW/EXAGIRL-7.8B-Instruct | finetune | — | 7.82B | M | other | huggingface-KREW | 6 | 686 |
| mlx-community/EXAONE-3.5-7.8B-Instruct-3bit | quantized | — | 7.82B | M | other | mlx-community | 0 | 959 |
| mobilint/EXAONE-3.5-7.8B-Instruct-Batch16 | quantized | — | 419M | S | other | mobilint | 0 | 901 |
| Jongbin-kr/llama3_convfin_lofit_top32_3ep | finetune | — | 7.82B | M | — | Jongbin-kr | 0 | 899 |
| DevQuasar-6/LGAI-EXAONE.EXAONE-3.5-7.8B-Instruct-GGUF | quantized | GGUF | — | — | — | DevQuasar-6 | 0 | 886 |
| huggingface-KREW/EXAONE-3.5-7.8B-Instruct-KoCulture | finetune | — | 7.82B | M | other | huggingface-KREW | 6 | 493 |
| Jongbin-kr/exaone-7b-verireason-reproduced-1513-fullft-epoch4 | finetune | — | 7.82B | M | other | Jongbin-kr | 0 | 724 |
| hymmmm/exaone-3.5-7.8b-emt-qlora-awq-v2 | quantized | awq | 7.82B | M | other | hymmmm | 0 | 722 |
| alexander583/EXAONE-3.5-7.8B-Instruct-Q4-mlx | quantized | mlx | 7.82B | M | other | alexander583 | 0 | 670 |
| Jongbin-kr/exaone-3.5-7.8b-verireason-full-ft-reasoning-co-ratio1.0-epoch3 | finetune | — | 7.82B | M | — | Jongbin-kr | 0 | 582 |
| Jongbin-kr/exaone3.5_lofit_convfinqa_top32 | finetune | — | 7.82B | M | — | Jongbin-kr | 0 | 561 |
| FloatDo/EXAONE-3.5-7.8B-Instruct-GGUF | quantized | GGUF | — | — | other | FloatDo | 0 | 520 |
| Jongbin-kr/exaone_7b_lora_rtlcoder_ratio1.0 | adapter | — | — | — | — | Jongbin-kr | 0 | 465 |
| furiosa-ai/EXAONE-3.5-7.8B-Instruct | finetune | — | — | — | other | furiosa-ai | 0 | 425 |
| werty1248/EXAONE-3.5-7.8B-Stratos-Ko | finetune | — | 7.82B | M | — | werty1248 | 1 | 323 |
| itlwas/EXAONE-3.5-7.8B-Instruct-Q4_K_M-GGUF | quantized | GGUF | — | — | other | itlwas | 0 | 325 |
| NikolayKozloff/EXAONE-3.5-7.8B-Instruct-Q8_0-GGUF | quantized | Q8_0 | — | — | other | NikolayKozloff | 1 | 264 |
| hoySky92/EXAONE-U-CURATE-Random1K-1epoch-v1 | finetune | — | 7.82B | M | other | hoySky92 | 0 | 301 |
| hymmmm/exaone-3.5-7.8b-emt-chatbot | finetune | — | 266K | S | other | hymmmm | 0 | 299 |
| Open4bits/EXAONE-3.5-7.8B-Instruct-GGUF | quantized | GGUF | — | — | other | Open4bits | 0 | 270 |
| Damisoda/EXAONE-Dementia-SAR | adapter | — | — | — | other | Damisoda | 5 | 18 |
| xxccho/EXAONE-3.5-7.8B-Instruct_lg_convfin_mcq_pc_lora_r64_DR1.0 | adapter | — | — | — | — | xxccho | 0 | 248 |
| xxccho/EXAONE-3.5-7.8B-Instruct_lg_convfin_mcq_reasoning_pc_lora_r64_DR1.0 | adapter | — | — | — | — | xxccho | 0 | 239 |
| alexgusevski/EXAONE-3.5-7.8B-Instruct-q3-mlx | quantized | mlx | 7.82B | M | other | alexgusevski | 0 | 234 |
Ranked by source signals (HN, Lobsters, authority domains, official posts) and recency. Summaries are generated from a stored copy of the article text; every sentence cites a span of that text, and nothing is shown that failed the citation check.
arXiv links found in the model card and its tags. "Peer-reviewed" is shown only with evidence: a published venue in OpenAlex (measured) or the authors' own journal reference on arXiv (reported).
Hub snapshot: 2026-10-25 · HF API: 2026-10-24 21:39 UTC · Providers: 2026-10-24 21:44 UTC · Articles: 2026-10-24 21:46 UTC · Summary: 2026-10-07 15:10 UTC · Last verified 2026-10-25
Email alerts are not open yet; the feeds above are the same data. See feeds for the item schema.
